Microsoft Foundry gives autonomous agents their own enterprise identities

Microsoft Foundry now lets enterprises build autonomous agents that operate in Teams and other Microsoft 365 applications under their own identities, rather than borrowing a user’s credentials. The agents can have their own accounts, permissions, audit trails and accountable managers.
- Foundry agents can work in Teams group chats and Microsoft 365 apps.
- Each agent has an Entra identity, user account, permissions and audit trail.
- Administrators approve blueprints before people create agent instances.
- Managers decide which business resources each agent instance can access.
- Agent 365 can manage the agent fleet, including a blueprint-level kill switch.
Agents designed to work with teams
Microsoft says the new approach addresses a limitation that has kept many enterprise agents focused on individual users. Without an account of their own, an agent working in Outlook or Teams must act with the permissions of a signed-in person. That creates problems when an event triggers an agent without a user present, or when an agent operates in a group conversation where it is unclear whose permissions should apply.
Foundry agents are designed to work in multi-user spaces such as Teams group chats and across Microsoft 365 applications. According to Microsoft, they can respond to email, edit Excel sheets and create PowerPoint decks, including when no one is signed in. Each agent acts as a named member of the organization, with a standing role rather than handling only one request at a time.
Each agent gets its own account and access
Every Foundry agent has a Microsoft Entra agent identity and an Entra agent user account. Microsoft says that account provides its own email, calendar, OneDrive, Teams presence and place in the organization chart. It also allows the agent to perform Microsoft 365 actions as itself.
Organizations can grant access in the same way they would for a person. An agent can receive access to calendars and mailboxes through Exchange delegation permissions, join security and Microsoft 365 groups, receive access packages and use shared documents. Its actions are attributed to the agent, rather than to the person who happened to interact with it.
Microsoft describes this as an accountability model. Every agent has a manager who is responsible for it, and its position in the organization chart shows who that manager is. The company also says the agents use existing Entra access controls and Agent 365 governance rather than requiring organizations to learn separate systems.
Blueprints separate capability from business access
Foundry agents are built as agent-as-a-service blueprints. A blueprint defines what an agent can do, including the APIs and MCP tools it can access, while custom code controls behaviors such as when it responds. After an administrator approves a published blueprint, approved people in the organization can create their own instances, each with a separate identity, account and access to resources.
Microsoft says capability and access remain separate. A release captain blueprint, for example, might know how to update Azure DevOps work items and write to SharePoint, while each manager decides which Azure DevOps project and SharePoint site its instance can reach.
The model also includes controls intended for agents’ distinct risks. Microsoft says one kill switch can disable an entire class of agents created from the same blueprint, while Agent 365 provides a control plane for managing the fleet. The company cautions that an agent can continue confidently beyond what it knows, and that multiple copies of one autopilot can share the same blind spots. For users, the change means autonomous agents can be placed into team workflows with identifiable owners, defined permissions and traceable actions, rather than operating only as extensions of individual accounts.
Microsoft Foundry: Build trusted autonomous agents for your team with Microsoft Foundry


